Review of Monte Carlo modeling of light transport in tissues - Caigang Zhu Quan Liu
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Review of Monte Carlo modeling of light transport in tissues Caigang Zhu Quan Liu Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Journal of Biomedical Optics 18(5), 050902 (May 2013) REVIEW Review of Monte Carlo modeling of light transport in tissues Caigang Zhu and Quan Liu Nanyang Technological University, School of Chemical and Biomedical Engineering, Division of Bioengineering, 637457 Singapore Abstract. A general survey is provided on the capability of Monte Carlo (MC) modeling in tissue optics while paying special attention to the recent progress in the development of methods for speeding up MC simulations. The principles of MC modeling for the simulation of light transport in tissues, which includes the general procedure of tracking an individual photon packet, common light–tissue interactions that can be simulated, frequently used tissue models, common contact/noncontact illumination and detection setups, and the treatment of time-resolved and fre- quency-domain optical measurements, are briefly described to help interested readers achieve a quick start. Following that, a variety of methods for speeding up MC simulations, which includes scaling methods, perturbation methods, hybrid methods, variance reduction techniques, parallel computation, and special methods for fluorescence simulations, as well as their respective advantages and disadvantages are discussed. Then the applications of MC methods in tissue optics, laser Doppler flowmetry, photodynamic therapy, optical coherence tomography, and diffuse optical tomography are briefly surveyed. Finally, the potential directions for the future development of the MC method in tissue optics are discussed. © The Authors. Published by SPIE under a Creative Commons Attribution 3.0 Unported License. Distribution or reproduction of this work in whole or in part requires full attribution of the original publication, including its DOI. [DOI: 10.1117/1.JBO.18.5.050902] Keywords: Monte Carlo; light transport in tissues; numerical simulation; tissue optics; optical spectroscopy. Paper 130036VR received Jan. 22, 2013; revised manuscript received Apr. 10, 2013; accepted for publication Apr. 15, 2013; published online May 10, 2013. 1 Introduction light–tissue interactions that can be simulated such as light absorption and scattering, frequently used tissue models, Monte Carlo (MC) methods are a category of computational common contact and noncontact illumination and detection set- methods that involve the random sampling of a physical quan- ups, and the treatment of time-resolved and frequency-domain tity.1,2 The term “the Monte Carlo method” can be traced back to optical measurements, are described in detail to help interested 1940s,1 in which it was proposed to investigate neutron transport readers achieve a quick start. Following that, a variety of meth- through various materials. Such a problem cannot be solved by ods for speeding up MC simulations, including scaling methods, conventional and deterministic mathematical methods. Due to perturbation methods, hybrid methods, variation reduction its versatility, this method has found applications in many differ- techniques, parallel computation, and special methods for fluo- ent fields3 including tissue optics. It has become a popular tool rescence simulations, and their respective advantages and disad- for simulating light transport in tissues for more than two dec- vantages are discussed. Then the biomedical applications of ades4 because it provides a flexible and rigorous solution to the MC methods, including the simulation of optical spectra, esti- problem of light transport in turbid media with complex struc- mation of optical properties, simulation of optical measurements ture. The MC method is able to solve radiative transport equa- in laser Doppler flowmetry (LDF), simulation of light dosage in tion (RTE) with any desired accuracy,5 assuming that the photodynamic therapy (PDT), simulation of signal source in required computational load is affordable. For this reason, optical coherence tomography (OCT) and diffuse optical tomog- this method is viewed as the gold standard method to model raphy (DOT), are surveyed. Finally, the potential directions for light transport in tissues, results from which are frequently used the future development of MC methods are discussed, which are as reference to validate other less rigorous methods such as dif- based on their current status in the literature survey and the fuse approximation to the RTE.6,7 Due to its flexibility and authors’ anticipation. It should be pointed out that this review recent advances in speed, the MC method has been explored is intended to give a general survey on the capability of MC in tissue optics to solve both the forward and inverse problems. modeling in tissue optics while paying special attention on In the forward problem, light distribution is simulated for given methods for speeding up MC simulations since the time- optical properties, whereas in the inverse problem, optical prop- consuming nature of common MC simulations could limit its erties are estimated by fitting the light distribution simulated by applications. the MC method to experimentally measured values. In this review paper, the principles of MC modeling for the simulation of light transport in tissues, including the general 2 Principles of MC Modeling of Light procedure of tracking an individual photon packet, common Transport in Tissues 2.1 General Procedure of Steady State MC Address all correspondence to: Quan Liu, Nanyang Technological University, Modeling of Light Transport in Tissues School of Chemical and Biomedical Engineering, Division of Bioengineering, Singapore 637457. Tel: 65-65138298; Fax: 65-67911761; E-mail: quanliu@ In the general procedure of MC modeling, light transport in tis- ntu.edu.sg sues is simulated by tracing the random walk steps that each Journal of Biomedical Optics 050902-1 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues photon packet takes when it travels inside a tissue model. For weight that, after traveling in the medium, escapes from the each launched photon packet, an initial weight is assigned as it same side of the tissue model as the incident light is scored enters the tissue model, as illustrated in Fig. 1. The step size will as diffuse reflectance. In contrast, the fraction of photon packet be sampled randomly based on the optical properties of the tis- weight that travels through the medium and escapes from the sue model. If it is about to hit a boundary, either of the following other side of the tissue model is scored as transmittance.5,8,9 two methods could be used to handle this situation. In the first To simulate fluorescence emission, one additional parameter, method, the photon packet will either transmit through or be which is fluorescence quantum yield,10,11 needs to be incorpo- reflected from the boundary. In the second method, a fraction rated to describe the probability that the absorbed photon packet of the photon packet’s weight will always be reflected and weight can be converted to a fluorescence photon at a different the remaining fraction of the photon packet’s weight will trans- wavelength. If time-resolved fluorescence is simulated, the life- mit through. The probabilities of transmission or reflection in time of fluorescence needs to be defined. The initial direction of the first method, and the fraction of the photon packet’s weight the fluorescence photon is isotropic due to the nature of fluores- transmitting through or being reflected in the second method, cence emission. As illustrated in Fig. 2, the MC modeling of are governed by Snell’s law and Fresnel’s equations. At the fluorescence propagation in tissues involves three steps.11–13 end of each step, the photon packet’s weight is reduced accord- The first step involves a general MC simulation to simulate light ing to the absorption probability; meanwhile, the new step size propagation with optical properties at the excitation wavelength. and scattering angle for the next step will be sampled randomly In the second step, a fluorescence photon may then be generated based on their respective probability distributions. The photon upon the absorption of an excitation photon with a probability packet propagates in the tissue model step by step until it exits defined by the quantum yield and time delay defined by the life- the tissue model or is completely absorbed. Once a sufficient time of fluorescence. The third step again involves a general number of photon packets are launched, the cumulative distri- MC simulation to simulate fluorescence light propagation with bution of all photon paths would provide an accurate approxi- optical properties at the emission wavelength. It is clear that mation to the true solution of the light transport problem and the simulated fluorescence from a tissue model will be related to contribution averaged from all photons can be used to estimate the absorption and scattering properties of the tissue model the physical quantities of interest. in addition to the fluorescence quantum yield and lifetime. Fluorescence simulation is typically much more time-consuming than the simulation of diffuse reflectance due to extra fluores- 2.2 Common Light–Tissue Interactions in MC Modeling cence photon propagation. To simulate Raman emission, a parameter similar to fluores- Several types of common light–tissue interactions, including cence quantum yield, named as Raman cross-section,14–17 is light absorption, elastic scattering, fluorescence and Raman needed to describe the probability that a Raman photon will scattering, have been simulated by the MC methods previously. be generated at each step. A phase function for Raman photons The absorption coefficient μa (unit: cm−1 ) and the scattering needs to be determined. The MC simulation procedure for coefficient μs (unit: cm−1 ) are used to describe the probability Raman light propagation will be similar to that for fluorescence. of absorption and scattering, respectively, occurring in a unit Bioluminescence refers to the phenomenon of living crea- path length. The anisotropy factor g, which is defined as the tures producing light, which results from the conversion of average cosine of scattering angles, determines the probability chemical energy to bioluminescence photons18 and which has distribution of scattering angles to the first-order approximation. In addition, the refractive index mismatch between any two regions in the tissue model or at the air–tissue interface will determine the angle of refraction. The fraction of photon packet Fig. 2 Flow chart for MC modeling of the propagation of a single photon packet, in which one set of wavelength change is involved. λexc indi- cates the excitation wavelength and λemm indicates the emission wave- Fig. 1 Flow chart for MC modeling of the propagation of a single photon length. The new photon packet with a different wavelength corresponds packet, in which no wavelength change is involved. to fluorescence or Raman light at a single emission wavelength. Journal of Biomedical Optics 050902-2 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues also been investigated using MC modeling. Because biolumi- transport in a living mouse model. The mouse model consists of nescence does not need an external light source for excitation, several segmented regions that are extended from several build- the first step in MC simulation of bioluminescence is to generate ing blocks such as ellipses, cylinders, and polyhedrons. This bioluminescence photons package according to the distribution platform is particularly suitable for small animal imaging. of bioluminescence sources.19,20 After that, the simulation of Margallo-Balbas et al.39 and Ren et al.40 have developed triangu- bioluminescence photon propagation in a tissue model is exactly lar-surface-based MC methods to model light transport in com- the same as the simulation of diffuse reflectance. plex tissue structures. The triangular-surface-based approach If the polarization property of light is considered in MC mod- allows an improved approximation to the interfaces between eling, the polarization of a photon can be represented by Stokes domains, but it is not able to model complex media with con- vectors and the polarimetry properties of the tissue model can be tinuously varying optical properties. Moreover, it could be described by Jones matrix or Mueller matrices,21,22 which will time-consuming to determine ray–surface intersection because a not be expanded in this review. range of triangles will have to be scanned. To overcome the lim- itations associated with triangular-surface-based MC method, 2.3 Common Tissue Models in MC Modeling most recently, Shen et al.41 as well as Fang42 have presented mesh-based MC methods, by which one can model much more Common tissue models used in MC simulations include the complex structures and situations. homogeneous and nonhomogeneous tissue models. The optical properties in a homogeneous tissue model are equal every- 2.4 Common Illumination and Detection Setups where.4,23 In contrast, the optical properties in a nonhomogene- in MC Modeling ous tissue model vary with the tissue region. The following survey is focused on nonhomogeneous tissue models because One important advantage of MC modeling, as compared to other of its high preclinical and clinical relevance. non-numerical methods such as diffuse approximation, is its The most commonly used nonhomogeneous tissue model is capability to faithfully simulate a variety of contact and noncon- perhaps the multilayered tissue model,8,9,12,13,24–30 which is fre- tact illumination and detection setups for optical measurements. quently employed to mimic epithelial tissues. In a multilayered Note that the contact setup requires the direct contact between tissue model, each tissue layer is assumed to be flat with uniform the tip of an optical probe and tissue samples. In contrast, the optical properties and it is infinitely large on the lateral dimen- noncontact setup enables optical measurements from a tissue sion. This assumption works fine when the source–detector sample without directly contacting it. separation is small so that the spatial variation in the optical properties within the separation is negligible. However, it could 2.4.1 Contact setup for illumination and detection cause significant errors if the optical properties change signifi- cantly in a small area, such as in dysplasia or early cancer31 and Fiber-optic probes are commonly used in contact illumination port wine stain (PWS) model.32 To overcome this limitation, tis- and detection configurations as demonstrated in many previous sue models including heterogeneities with well-defined shapes reports.43 In general, these fiber-optic probes could be divided have been used to mimic complex tissue structures from differ- into two groups. In the first group, the same fiber or fiber bundle ent organs. For example, Smithies et al.32 and Lucassen et al.33 is used for both illumination and detection,30,44,45 while in the independently proposed MC models in which simple geometric second group, separate fibers are used for illumination and shapes were incorporated into layered structures to model light detection.12,46–48 There is no difference in the treatment of these transport in PWS model. In their PWS models, infinitely long two groups of fiber-optic probes from the point of view of mod- cylinders were buried in the bottom dermal layer to mimic blood eling because the first group of probes can be viewed as two vessels. Wang et al.34 reported an MC model in which a sphere separate and identical fibers or fiber bundles for illumination was buried inside a slab to model light transport in human and detection that happen to locate at the same spatial position. tumors. Zhu et al.31,35 proposed an MC model in which cuboid The key parameters in simulated fiber-optic probes include tumors were incorporated into layered tissues to model light the radii, numerical apertures (NA), tilt angles of illumination transport in early epithelial cancer models including both squ- and detection fibers, and the center-to-center distance between amous cell carcinoma and basal cell carcinoma. the two sets of fibers (which is called the source–detector sep- Voxelated tissue models have been also explored to simulate aration), as well as the refractive indices of these fibers relative irregular structures. Pfefer et al.36 reported a three-dimensional to that of the tissue model. The radius and NA of the illumina- (3-D) MC model based on modular adaptable grids to model tion fiber in combination with the radial and angular distribu- light propagation in geometrically complex biological tissues tions of photons coming out of the fiber define the locations and validated the code in a PWS model. Boas et al.37 proposed and the incident angles of incident photons. For a commonly a voxel-based 3-D MC model to model arbitrary complex used multimode fiber, the spatial locations and incident angles tissue structures and tested the code in an adult head model. of launched photons are typically assumed to follow uniform Patwardhan et al.38 also proposed a voxel-based 3-D MC code distribution and Gaussian distribution. Both spatial locations for simulating light transport in nonhomogeneous tissue struc- and incident angles need to undergo spatial coordinate transfor- tures and tested the code in a skin lesion model. The three voxel- mation when the tilt angle of the illumination fiber is larger than based MC codes above showed great flexibility in a range of zero. Here the tilt angle of a fiber refers to the angle of the fiber applications. However, to model tissue media with curved boun- axis relative to the normal axis of the tissue model. The incident daries in a voxel-based MC model, the grid density will have to beam could be also assumed to be collimated or focused. be increased, which requires more memory and computation. A Light detection by a fiber usually contains two steps. The few other approaches have been explored to accommodate this first step is to determine whether an exiting photon could enter situation. Li et al.19 reported a public MC domain, named mouse the area defined by the radius of the detection fiber. If it is true, optical simulation environment, to model bioluminescent light the second step is to determine whether the exiting direction of Journal of Biomedical Optics 050902-3 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues the photon falls within the acceptable angle of the detection compared to other analytical or empirical methods. Significant fiber calculated from the NA and refractive index of the fiber. efforts have been made to speed up the MC simulation of If the tilt angle of the detection fiber is larger than zero, the light transport in tissues during the past decades. These accel- exiting location and angle are subject to spatial coordinate eration methods can be roughly divided into several categories transformation. as follows. 2.4.2 Noncontact setup for illumination and detection 3.1 Scaling Methods Noncontact setups usually employ various lenses for illumina- tion and detection. In these setups, an adjunct lens or a combi- A typical scaling method requires a single or a few baseline MC nation of lenses is usually placed between a fiber-optic probe simulations, in which the histories of survival photons such as and the tissue sample to achieve noncontact measurements trajectories or step sizes are recorded. Then, diffuse reflectance while maintaining the well-defined illumination and detection or transmittance for a tissue model with different optical proper- geometry. Jaillon et al.49 proposed a method to simulate a bev- ties can be estimated by applying scaling relations on the eled fiber-optic probe coupled with a ball lens to achieve depth- recorded photon histories. These methods take advantage of sensitive fluorescence measurements from layered tissue mod- the fact that the scattering properties determine photon paths els. Later, the same group incorporated a half-ball lens into the and the absorption property only influences the weights of sur- beveled fiber-optic probe to achieve the same purpose with a vival photons. Graaff et al.68 proposed a limited scalable MC higher sensitivity.50 Zhu et al.51 proposed a method to simulate method for fast calculation of total reflectance and transmittance a fiber-optic probe coupled with convex lenses to achieve non- from slab geometries with different optical properties. It was contact depth-sensitive diffuse reflectance measurements from demonstrated that the trajectory information obtained in a refer- early tumors in an epithelial tissue model. By manipulating the ence MC simulation with a known albedo, i.e., μs ∕ðμa þ μs Þ, lens combination, an ordinary cone configuration and a special can be used to find the total reflectance and total transmittance cone shell configuration were investigated. It was found that the from slabs with other albedos. Kienle et al.69 extended Graaff’s cone shell configuration provides higher depth sensitivity to the theory to simulate space- and time- resolved diffuse reflectance tumor than the cone configuration. from a semi-infinite homogeneous tissue model with arbitrary optical properties. Their approach was based on scaling (for dif- ferent scattering coefficients) and re-weighting (for different 2.5 Time-Resolved and Frequency-Domain MC absorption coefficients) a discrete representation of the diffuse Modeling reflectance from one baseline MC simulation in a nonabsorbing Time-resolved optical measurements such as fluorescence life- semi-infinite medium. It is powerful, but both the discrete rep- time imaging (FLIM)52 and the complementary frequency- resentation and interpolation could introduce errors that are domain measurements such as frequency domain photon migra- often amplified in scaling. Pifferi et al.70 proposed a similar tion have received increasing attention recently, which have also approach to estimate space- and time-resolved diffuse reflec- been investigated in MC modeling. A time-domain technique tance and transmittance from a semi-infinite homogeneous usually measures the temporal point spread function (PSF) or the tissue model with arbitrary optical properties. Different from spreading of a propagating pulse in time.53,54 A frequency-domain Kienle’s method, the evaluation of reflectance and transmittance technique measures the temporal modulation transfer function in Pifferi’s approach is based on interpolation of results from or the attenuation and phase delay of a periodically varying pho- MC simulations for a range of different scattering coefficients, ton density wave.55,56 The two techniques are related by Fourier and scaling is performed for absorption coefficients. This transform. Several groups have developed time-domain MC approach increases the accuracy of results for different scatter- models37,57–60 and frequency-domain MC models61–65 to simulate ing coefficients at the cost of a significantly increased number of light transport in tissue. In the MC simulation of time-resolved baseline MC simulations. measurements, all the steps are the same as in steady-state The methods reviewed above are fast, but the binning and measurements, except that one additional parameter, i.e., interpolation involved introduce errors. In order to improve the time, is used to keep track of the time at which each event accuracy of these methods, Alerstam et al.60 improved Kienle’s occurs.37,57–60 The refractive index in each tissue region will in- method by applying scaling to individual photons. In this method, fluence the time that photons take to travel through. In the sim- the radial position of the exiting location and the total path length ulation of FLIM, it needs to be pointed out that the time delay of each detected photon are recorded and the trajectory informa- from photon absorption to fluorescence generation should fol- tion of each photon will be individually processed to find the sur- low the probability density distribution defined by the fluores- vival photon weight for tissue media with other sets of optical cence lifetime.66,67 In the frequency-domain measurements, the properties. Martinelli et al.71 derived a few scaling relationships modulation and/or phase delay of detected waves were ana- from the RTE, and their derivation showed that a rigorous appli- lyzed. The modulation and phase delay can be simulated in cation of the scaling method requires rescaling to be performed either a direct approach64 or an indirect approach, i.e., using for each photon’s biography individually. Two basic relations for Fourier transformation from a time-domain MC simulation.65 scaling a survival photon’s exit radial position r and exit weight w are listed in Eqs. (1) and (2) below.47 3 Methods for the Acceleration of MC Simulation μt r0 ¼ r ⋅ ; (1) While the MC method is the gold standard method to model μt0 light transport in turbid media, the major drawback of the MC method is the requirement of intensive computation to achieve 0 N 0 α results with desirable accuracy due to the stochastic nature of w ¼w⋅ ; (2) MC simulations, which makes it extremely time-consuming α Journal of Biomedical Optics 050902-4 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues in which r, w, μt , and α are the exit radial position, exit weight, transport to estimate optical properties in the perturbed region transport coefficients, and albedo in the baseline simulation, while as surveyed below. r 0, w 0 , μt0 , and α 0 are those in the new simulations. N is the num- Sassaroli et al.75 proposed two perturbation relations to esti- ber of collisions recorded in the baseline simulation before the mate the temporal response in diffuse reflectance from a photon exits. Two relations essentially assume that the same set medium, in which scattering or absorbing inhomogeneities of random numbers sampled in the baseline simulation are also are introduced, from the trajectory information obtained from used in the new simulation and everything remains unchanged in the baseline simulation of a homogeneous medium. Hayakawa two simulations, except the absorption and scattering coefficients. et al.76 demonstrated that the perturbation relation can be Illumination and detection geometries have also been incor- directly incorporated into a two-parameter Levenberg-Marquardt porated into the scaling procedure. Palmer et al.47 extended algorithm to solve the inverse photon migration problems in a Graaff’s scaling method from illumination by a pencil beam two-layered tissue model rapidly. Recently, the same group77 to that by an optical fiber, and they also extended the original demonstrated the use of this method for extraction of optical scaling method from the total reflectance to the reflectance properties in a layered phantom mimicking an epithelial tissue detected by an optical fiber by combing scaling and convolution. model for given experimental measurements of spatially Wang et al.72 proposed two convolution formulas for the scaling resolved diffuse reflectance. This method was found effective MC method to calculate diffuse reflectance from a semi-infinite over a broad range of absorption (50% to 400% relative to medium for a single illumination–detection fiber. Nearly all the the baseline value) and scattering (70% to 130% relative to previous papers about scaling dealt only with a homogeneous the baseline value) perturbations. However, this method requires tissue model. Liu et al.73 developed a method that applies the both the thickness of the epithelial layer and the optical proper- scaling method to multilayered tissue models. In this method, ties of one of the two layers. the homogeneous tissue model in a single baseline MC simu- Many other groups also proposed pMC-based methods for lation is divided into multiple thin pseudo layers. The horizontal the recovery of the optical properties in various tissue models. offset and the number of collisions that each survival photon Kumar et al.78 have presented a pMC-based method for recon- experienced in each pseudo layer are recorded and used later structing the optical properties of a heterogonous tissue model to scale for the exit distance and exit weight of the photon in with low scattering coefficients and the method was validated a multilayered tissue model with different set of optical proper- experimentally.29 Their results show that a priori knowledge ties. The method has been validated on both two-layered and of the location of inhomogeneities is important to know in three-layered epithelial tissue models. the reconstruction of optical properties of a heterogeneous tis- sue. More recently, Sassaroli et al.79 proposed a fast pMC method for photon migration in a tissue model with an arbitrary 3.2 Perturbation MC Methods distribution of optical properties. This method imposes a min- imal requirement on hard disk space; thus it is particularly suit- Similar to the scaling method, the perturbation MC (pMC) able to solve inverse problems in imaging, such as DOT. Zhu et method requires one baseline simulation in which the optical al.35 proposed a hybrid approach combining the scaling method properties are supposed to be close to the optical properties and the pMC method to accelerate the MC simulation of diffuse in the new tissue model so that the approximation made by per- reflectance from a multilayered tissue model with finite-size turbation is valid.74 The trajectory information including the exit tumor targets. Besides the advantage in speed, a larger range weight, path length, and number of collisions of each detected of probe configurations and tumor models can be simulated photon spent in the region of interest will be recorded in the by this approach compared to the scaling method or the pMC baseline simulation. Then the relation between the survival method alone. weight in the baseline simulation and that in the new tissue model based on the perturbation theory,75,76 i.e., 3.3 Hybrid MC Methods 0 j μ Hybrid MC methods incorporate fast analytical calculations wnew ¼ w ⋅ s ⋅ exp½−ðμt0 − μt ÞS; (3) μs such as diffuse approximation into a standard MC simulation. Flock et al.80 proposed a hybrid method to model light distribu- is used to estimate diffuse reflectance from the tissue model in tion in tissues. In this model, a series of MC simulations for which the optical properties of the interesting region are per- multiple sets of optical properties and geometrical parameters turbed. In Eq. (3), w, μs , and μt are the exit weight, scattering were performed to create a coupling function. Then, this cou- coefficient, and transport coefficient in the baseline simulation, pling function was used to correct the results computed by dif- while wnew , μs0 , and μt0 are those in the new simulation. S and j fusion theory. Wang et al.81 proposed a conceptually different are the photon path length and the number of collisions that a hybrid method to simulate diffuse reflectance from semi-infinite detected photon experienced in the perturbed region, respec- homogeneous media. Wang’s method combined the strength of tively, recorded in the baseline simulation. It should be pointed MC modeling in accuracy at locations near the light source and out that the pMC is an approximation in nature, so its accuracy the strength of diffusion theory in speed at locations distant from depends on the magnitude of difference in the optical properties the source. Wang et al.82 later extended this method from semi- between the perturbed optical properties in the new tissue model infinite media to turbid slabs with finite thickness, which is more and the original optical properties in the baseline simulation. useful than the previous method in practice. Alexandrakis et al.62 In contrast, the scaling method is precise in nature regardless proposed a fast diffusion-MC method for simulating spatially of the differences in optical properties because no approxima- resolved reflectance and phase delay in a two-layered human tion is made in scaling. One important advantage of the pMC is skin model, which facilitates the study of frequency-domain its simplicity and fast speed when the perturbed region is small, optical measurements. This method has been proven to be therefore it has been explored in the inverse problem of light several hundred times faster than a standard MC simulation. Journal of Biomedical Optics 050902-5 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues Hayashi et al.83 presented a hybrid method to model light propa- other volumes. Chen et al.90 proposed a controlled MC method gation in a head model that contains both high-scattering regions in which an attractive point with an adjustable attractive factor and low-scattering regions. Light propagation in high-scattering was introduced to increase the efficiency of trajectory generation regions was calculated by diffusion approximation and that in by forcing photons to propagate along directions more likely to the low-scattering region, i.e., the cerebrospinal fluid layer, was intersect with the detector, which is similar to geometry splitting simulated by the MC method. Since the time-consuming MC in principle. They first demonstrated this approach in transmis- simulation is employed only in part of the head model, the com- sion geometry90 and then in reflection geometry.91 Behin-Ain putation time is significantly shorter than that of the standard et al.92 extended Chen’s method for the efficient construction MC method. Donner et al.84 presented a diffusion-MC method of the early temporal PSF created by the visible or near-infrared for fast calculation of steady-state diffuse reflectance and trans- photons transmitting through an optically thick scattering mittance from layered tissue models. In their method, the medium. More recently, Lima et al.93,94 incorporated an steady-state diffuse reflectance and transmittance profiles of improved importance sampling method into a standard MC for each individual layer were calculated and then convolved to gen- fast MC simulation of time-domain OCT, by which several hun- erate the overall diffuse reflectance and transmittance to elimi- dred times of acceleration has been achieved. nate the need of considering boundary conditions. Luo et al.85 introduced an improved diffusion model derived empirically. 3.5 Parallel Computation-Based MC Methods Then the modified diffusion model was combined with the MC method to estimate diffuse reflectance from turbid media with a Parallel computation has received increasing attention recently high ratio of the absorption coefficient to the reduced scattering in the study of speeding up MC simulations due to advances in coefficient, which can be as large as 0.07. Di Rocco et al.86 pro- computer technology. The acceleration due to parallel compu- posed a hybrid method to speed up MC simulations in slab tation is independent of all previous techniques and thus could geometries including deep inhomegeneities. In this approach, be used in combination with them to gain extra benefit. Kirkby the tissue model was treated as two sections, i.e., the top et al.95 reported an approach by which one can run an MC sim- layer with a thickness of d in which there is no inhomogeneity ulation simultaneously on multiple computers, aiming to utilize and the bottom layer with inhomgeneity. Propagation up to the the unoccupied time slots of networked computers to speed up given depth d, i.e., the top layer, is replaced by analytical cal- MC simulations. This method has reduced simulation time culations using diffusion approximation. Then photon propaga- appreciably. However, it can be time-consuming to wait for all tion is continued inside the bottom layer using MC rules until computers to update the result files in order to get the final the photon is terminated or detected. Tinet et al.58 adapted the result. Moreover, the requirement of saving disk space imposes statistical estimator technique used previously in the nuclear the use of binary files, and this raised compatibility issues across engineering field to a fast semi-analytical MC model for simu- in various types of computers. Colasanti et al.96 explored a dif- lating time-resolved light scattering problems. There were two ferent approach to address the limitations associated with steps in this approach. The first step was information generation, Kirkby’s method. They developed an MC multiple-processor in which the contribution to the overall reflectance and transmit- code that can be run on a computer with multiple processors tance was evaluated for each scattering event. The second step instead of running on many single-processor computers. The was information processing, in which the results of first step results showed that the parallelization reduced computation were used to calculate desired results analytically. Chatigny time significantly. et al.87 proposed a hybrid method to efficiently model the time- Considerable efforts have also been made to implement MC and space-resolved transmittance through a breast tissue model codes in graphics processing unit (GPU) environment to speed that was divided into multiple isotropic regions and anisotropic up MC simulations. Erik et al.97 proposed a method that was regions. In this hybrid method, the standard MC method incor- executed on a low-cost GPU to speed up the MC simulation porated with the isotropic diffusion similarity rule was applied of time-resolved photon propagation in a semi-infinite medium. to the area that contains both isotropic and anisotropic regions, The results showed that GPU-based MC simulations were while the analytical MC, which is similar to Tinet’s method, was 1000 times faster than those performed on a single standard cen- used for the area that contains isotropic regions only. tral processing unit (CPU). The same group98 further proposed an optimization scheme to overcome the performance bottle- 3.4 Variance Reduction Techniques neck caused by atomic access to harness the full potential of GPU. Martinsen et al.99 implemented the MC algorithm on In addition to hybrid methods reviewed above, multiple variance an NVIDIA graphics card to model photon transport in turbid reduction techniques, which were initially applied in modeling media. The GPU-based MC method was found to be 70 times neutron transport,88 have also been investigated in the MC mod- faster than a CPU-based MC method on a 2.67 GHz desktop eling of light transport in tissues. For example, the weighted computer. Fang et al.100 reported a parallel MC algorithm accel- photon model and Russian roulette scheme have been employed erated by GPU for the simulation of time-resolved photon in the public-domain MC code, Monte Carlo modeling of pho- propagation in an arbitrary 3-D turbid media. It has been dem- ton transport in Multi-Layered tissues.8 Liu et al.89 have used onstrated that GPU-based approach was 300 times faster than one of the oldest and the most widely used variance reduction the conventional CPU approach when 1792 parallel threads techniques in MC modeling, i.e., geometry splitting, to speed up were used. Ren et al.40 presented an MC algorithm that was the creation of an MC database to estimate the optical properties implemented into GPU environment to model light transport of a two-layered epithelial tissue model from simulated diffuse in a complex heterogeneous tissue model in which the tissue reflectance. In this strategy, the tissue model is separated into surface was constructed by a number of triangle meshes. The several volumes, and the technique can reduce variances in cer- MC algorithm has been tested and validated in a heterogeneous tain important volumes by increasing the chance of sampling in mouse model. Leung et al.101 proposed a GPU-based MC model important volumes and decreasing the chance of sampling in to simulate ultrasound modulated light in turbid media. It was Journal of Biomedical Optics 050902-6 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues found that a GPU-based simulation was 70 times faster com- a convolution-based MC method to accelerate the simulation of pared to CPU-based approach on the same tissue model. Most fluorescence spectra from layered tissues. Their method recently, Cai et al.102 implemented a fast perturbation MC exploited the symmetry property of the problem, which requires method proposed by Angelo79 on GPU. It has been demon- the multilayered tissue model to be infinite in the radial dimen- strated that the GPU-based approach was 1000 times faster com- sion. Different from the conventional fluorescence MC code, pared to the conventional CPU-based approach. this method computed the excitation and emission light profiles Besides using GPU to speed up the MC simulations, some separately, from which the spatial distribution of absorption and researchers have explored using field-programmable gate arrays emission probabilities were obtained. Then a convolution (FPGA) to accelerate MC simulations. For example, Lo et al.103 scheme will be applied on the absorption probability and emis- implemented an MC simulation on a developmental platform sion probability data to get the final fluorescence signals. with multiple FPGAs. The FPGA-based MC simulation was Swartling’s method has been used by Palmer et al.113,114 to cre- found to be 80 times faster and 45 times more energy efficient, ate an MC database for fluorescence spectroscopy to estimate on average, than the MC simulation executed on a 3 GHz Intel the fluorescence property of a breast tissue model from fluores- Xeon processor. cence measurement using a fiber-optic probe. Liebert et al.57 Recently, Internet-based parallel computation has gained developed an MC code for fast simulation of time-resolved fluo- increasing attention for fast MC modeling of light transport rescence in layered tissues. In this method, both the spatial dis- in tissues. Pratx et al.104 reported a method for performing tribution of fluorescence generation and the distribution of times MC simulation in a massively parallel cloud computing environ- arrival (DTA) of fluorescence photons at the detectors were cal- ment based on MapReduce developed by Google. For a cluster culated along the excitation photons’ trajectories. Then the dis- size of 240 nodes, an improvement in speed of 1258 times was tribution of fluorescence generation inside the medium and DTA achieved as compared to the single threaded MC program. as well as the fluorescence conversion probability were used to Doronin et al.105 developed a peer-to-peer (P2P) MC code to calculate the final fluorescence signal. It should be noted that the provide multiuser access for the fast online MC simulation of reduced scattering coefficients at the excitation and emission photon migration in complex turbid media. Their results showed wavelengths have to be approximately equal in this method. that this P2P-based MC simulation was three times faster than the GPU-based MC simulations. 3.7 Comparison of Methods for MC Acceleration Most methods surveyed in the previous sections have been 3.6 Acceleration of MC Simulation of Fluorescence compared and summarized in Table 1 with respect to their accel- The methods reviewed above are all about the acceleration of eration performance, relative error in simulated optical measure- MC simulation of diffuse reflectance or transmittance. Compared ments, respective advantages, and limitations. It should be noted to diffuse reflectance, fluorescence simulation is more complex that those parallel computation-based methods were not listed in and much more time-consuming due to the generation of fluo- this table because its performance highly depends on the com- rescence photons upon each absorption event of an excitation puting architecture, and all the methods summarized in this table photon. A number of groups11–13,30,57,106–108 have employed MC can be further accelerated by applying parallel computation. modeling to simulate fluorescence in tissues due to the growing interest in fluorescence spectroscopy or imaging for medical 4 Applications of MC Methods in Tissue Optics applications.109–112 As a consequence, multiple groups have The most common application of MC method in tissue optics is investigated various methods to speed up the MC simulation the simulation of optical measurements such as diffuse reflec- of fluorescence in biological tissues. Swartling et al.13 proposed tance, transmittance, and fluorescence for a given tissue Table 1 Comparison of various methods in MC acceleration. Relative error in Acceleration relative simulated optical Methods to standard MC measurements Advantages Limitations Scaling MC ∼200 (Ref. 73) Less than 4% (Ref. 73) No approximation is made, Applicable to layered tissue and it is accurate and fast. models only so far. Perturbation MC ∼1300 (Ref. 79) Can be less than 4% It is applicable to tissue Sensitive to perturbation in depending on the with complex structures. scattering properties. magnitude of perturbation (Ref. 79) Hybrid MC ∼300 (Ref. 82) Around 5% (Ref. 82) It has a larger applicable Relatively complicated range than pMC. computation. The particular region has to be homogeneous. Variance ∼300 (Refs. 93 and 94) Around 5% (Refs. 93 There are a variety of Limitation varies with the reduction and 94) choices available. specific technique. Note: The improvement relative to standard MC was defined as the fold of improvement in computation speed compared to a standard MC simulation in order to obtain results with comparable variance. GPU-based methods were not listed in this table because all the methods summarized in this table can be further accelerated by GPU. Journal of Biomedical Optics 050902-7 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues model and illumination/detection geometry, which is considered lower150,151 than 10−7 . Different from that, coherent Raman as a forward problem. In this situation, MC simulations could techniques utilize laser beams at two different frequencies to provide guidelines for the selection of optimal illumination/ produce a coherent output, which result in much stronger coher- detection geometry for selective optical measurements.46,49,115–118 ent Raman signals compared to spontaneous Raman scattering. In contrast, MC simulations can also provide data to estimate Because of the high chemical specificity of Raman spectros- the optical properties of a tissue model from optical measure- copy, it is anticipated that there will be more studies using ments, which is considered as an inverse problem. Solving an the MC method for Raman spectroscopy to optimize experimen- inverse problem typically involves the use of a nonlinear least tal setup. One important issue in these studies is that, the phase square error algorithm47,89 or a similar algorithm to find a set of function of Raman scattering from biological components in tis- optical properties that would yield optical measurements in MC sues have not been systematically studied. Recent MC studies simulations best matching the actual measurements. Due to the on Raman scattering14,15 assumed isotropic Raman emission. slow speed of traditional MC simulations, a database is frequently This assumption should work fine for spontaneous Raman created a priori in such an inverse problem to speed up the inver- scattering according to a numerical study.152 However, this sion process.119 Most of the acceleration methods discussed assumption is not valid for coherent Raman scattering since above can be employed in the creation of such an MC database. the angular distribution of Raman emission is affected by The MC method has been frequently used to find the optimal both the wavelength of the pump light source and the propagat- optical configuration in LDF, one of the oldest techniques in ing beam geometry.152–154 A systematic study on the phase func- biomedical optics during the past decade. Jentink et al.120,121 tion of Raman scattering on the molecule level for Raman active used MC simulations to investigate the relationship between biological molecules such as protein and DNA and on the sub- the output of laser Doppler perfusion meters and the optical cellular level for organelles such as mitochondria will be very probe configuration as well as the tissue scattering properties. helpful, in which one or a couple of key parameters similar to Stern et al.122 used MC modeling to simulate the spatial Doppler the anisotropy factor in elastic scattering could accurately sensitivity field of a two-fiber velocimeter, by which an optimal describe the angular distribution of Raman scattering in most fiber configuration was identified. Similar applications can also common cases. The use of such validated phase functions in be found in Refs. 123 through 125. Recently, MC method has MC simulations will yield more useful information than the sim- been incorporated into LDF to estimate blood flow126,127 or the plistic treatment in the current literature. phase function of light scattering.128 The MC method plays an important role in the selection of 5.2 Incorporation of More Realistic Elastic Light optimal configuration for PDT because it can generate light dis- Scattering Model into the MC Method tribution in a complex tissue model for PDT dosage determina- tion. Barajas et al.129 simulated the angular radiance in tissue Despite the exploration of various inhomogeneous tissue models phantoms and human prostate model to characterize light discussed above, including the multilayered tissue model, voxel- dosimetry using the MC method. Liu et al.130 used the MC based and mesh-based tissue models, these tissue models are all method to simulate the temporal and spatial distributions of based on a few simple optical coefficients including the scatter- ground-state oxygen, photosensitizer, and singlet oxygen in a ing coefficients and anisotropy factor to characterize optical skin model for the treatment of human skin cancer. Valentine scatterers. A complete phase function could be used to provide et al.131 simulated in vivo protoporphyrin IX (PpIX) fluores- the comprehensive information related to the morphology of cence and singlet oxygen production during PDT for patients optical scatterers, but it is inconvenient for use and its physical with superficial basal cell carcinoma. Later, the same group132 meaning is not straightforward. From these scattering proper- used the MC method to identify optimal light delivery configu- ties, the scatterer size and density can be derived47,155 if they ration in PDT on nonmelanoma skin cancer. are assumed to be uniformly distributed spheres with homo- The MC method has also been investigated to simulate the geneous density. In many scenarios, these assumptions are OCT signals133,134 and images135–137 during past years due to its not valid. For example, it is commonly known that the size flexibility and high accuracy. Moreover, with the development and shapes of cells vary significantly with the depth from the of efficient MC methods, researchers have started to explore the tissue surface, and they also change with carcinogenesis. MC method for image reconstruction in DOT.138,139 From this point of view, the superposition of multiple phase functions156 or the fractal distribution of the scatterer size157 5 Discussion on the Potential Future Directions have been proposed to accommodate special situations. An Due to advances in computing technology, it is expected that the equiphase-sphere approximation for light scattering has also applications of the MC method will be expanded in the near been proposed by Li et al.158 to model inhomogeneous micro- future. A few potential directions in the development of the particles with complex interior structures. Later, the same group MC method are discussed below. reported two stochastic models,159 i.e., the Gaussian random sphere model and the Gaussian random field model, to simulate 5.1 Phase Function of Raman Scattering irregular shapes and internal structures in tissues. The incorpo- ration of these more realistic elastic light scattering models into Raman spectroscopy has been explored extensively for tissue the MC method will expand its capability and offer more accu- characterization15,17,140,141 including cancer diagnosis.14,142–149 rate information about light scatterers in tissues. Depending on whether the excitation light is coherent or inco- herent, Raman scattering can be broken down into two catego- 5.3 Exploration of the MC Method in Imaging ries, i.e., spontaneous Raman scattering or coherent Raman Reconstruction scattering. The signal generated out of spontaneous Raman scattering is typically very weak, in which the probability of In most current applications of the MC method, the tissue model generating a Raman photon for every excitation photon is is assumed to be a simple layered model or determined a priori, Journal of Biomedical Optics 050902-8 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
Zhu and Liu: Review of Monte Carlo modeling of light transport in tissues which does not fully exploit the potential of the MC method in 10. A. J. Welch and R. Richards-Kortum, “Monte Carlo simulation of propa- preclinical or clinical imaging/spectroscopy. When the MC gation of fluorescent light,” in Laser-Induced Interstitial Thermotherapy, G. Muller and A. Roggan, Eds., pp. 174–189, SPIE, Bellingham, method becomes adequately fast in the near future, which might Washington (1995). be mostly attributed to the combination of the accelerated MC 11. A. J. Welch et al., “Propagation of fluorescent light,” Lasers Surg. Med. methods and parallel computing discussed above, the MC 21(2), 166–178 (1997). method could be used to reconstruct the optical properties of 12. Q. Liu et al., “Experimental validation of Monte Carlo modeling of a complex tissue in optical tomography, in which the morpho- fluorescence in tissues in the UV-visible spectrum,” J. Biomed. Opt. logical structure could be obtained in real time by fast imaging 8(2), 223–236 (2003). techniques such as OCT for superficial regions of interest or 13. J. Swartling et al., “Accelerated Monte Carlo models to simulate fluo- rescence spectra from layered tissues,” J. Opt. Soc. Am. A 20(4), magnetic resonance imaging for deep regions of interest in a 714–727 (2003). large tissue volume. Those voxelated tissue models,19,36–42 14. M. D. Keller et al., “Monte Carlo model of spatially offset Raman spec- which are more realistic than simplified homogeneous or lay- troscopy for breast tumor margin analysis,” Appl. Spectrosc. 64(6), ered tissue models, can be readily used in such reconstruction. 607–614 (2010). The speed of reconstruction might be comparable to that using 15. C. Reble et al., “Quantitative Raman spectroscopy in turbid media,” diffusion approximation reported in the current literature, but the J. Biomed. Opt. 15(3), 037016 (2010). accuracy would be considerably higher in a tissue model on the 16. A. M. K. Enejder et al., “Blood analysis by Raman spectroscopy,” Opt. Lett. 27(22), 2004–2006 (2002). millimeter scale. 17. W. C. Shih et al., “Intrinsic Raman spectroscopy for quantitative bio- logical spectroscopy Part I: theory and simulations,” Opt. Express 6 Conclusion 16(17), 12726–12736 (2008). 18. J. W. Hastings, “Biological diversity, chemical mechanisms, and the In this review, the principles of MC modeling for the simulation evolutionary origins of bioluminescent systems,” J. Mol. Evol. 19(5), of light transport in tissues were described at the beginning. 309–321 (1983). Then a variety of methods for speeding up MC simulations were 19. H. Li et al., “A mouse optical simulation environment (MOSE) to inves- discussed to overcome the time-consuming weakness of MC tigate bioluminescent phenomena in the living mouse with the Monte modeling. Then the applications of MC methods in biomedical Carlo method,” Acad. Radiol. 11(9), 1029–1038 (2004). 20. D. Kumar et al., “Monte Carlo method for bioluminescence tomogra- optics were briefly surveyed. Finally, the potential directions for phy,” Indian J. Exp. Biol. 45(1), 58–63 (2007). the future development of the MC method in tissue optics were 21. J. C. Ramella-Roman et al., “Three Monte Carlo programs of polarized discussed. We hope this review has achieved its intended pur- light transport into scattering media: part I,” Opt. Express 13(12), 4420– pose to give a general survey on the capability of MC modeling 4438 (2005). in tissue optics and other relevant key techniques for those read- 22. J. C. Ramella-Roman et al., “Three Monte Carlo programs of polarized ers who are interested in MC modeling of light transport. light transport into scattering media: part II,” Opt. Express 13(25), 10392–10405 (2005). 23. J. M. Maarek et al., “A simulation method for the study of laser trans- Acknowledgments illumination of biological tissues,” Ann. Biomed. Eng. 12(3), 281–304 (1984). The authors would like to acknowledge the financial support 24. P. Van der Zee and D. T. Delpy, “Simulation of the point spread function from Tier 1 grant and Tier 2 grant funded by the Ministry of for light in tissue by a Monte Carlo method,” Adv. Exp. Med. Biol. 215, Education in Singapore (Grant Nos. RG47/09 and MOE 179–191 (1987). 2010-T2-1-049). 25. S. Avrillier et al., “Monte-Carlo simulation of collimated beam trans- mission through turbid media,” J. Phys. France 51(22), 2521–2542 (1990). References 26. Y. Hasegawa et al., “Monte-Carlo simulation of light transmission through living tissues,” Appl. Opt. 30(31), 4515–4520 (1991). 1. N. Metropolis and S. Ulam, “The Monte Carlo method,” J. Am. Stat. 27. G. Zaccanti, “Monte-Carlo study of light-propagation in optically thick Assoc. 44(247), 335–341 (1949). media-point-source case,” Appl. Opt. 30(15), 2031–2041 (1991). 2. I. Lux and L. Koblinger, Monte Carlo Particle Transport Methods: 28. H. Key et al., “Monte-Carlo modeling of light-propagation in breast-tis- Neutron and Photon Calculations, CRC Press, Boca Raton, Florida sue,” Phys. Med. Biol. 36(5), 591–602 (1991). (1991). 29. P. K. Yalavarthy et al., “Experimental investigation of perturbation 3. M. E. J. Newman and G.T. Barkema, Monte Carlo Methods in Monte-Carlo based derivative estimation for imaging low-scattering tis- Statistical Physics, Clarendon Press, Oxford University Press, Oxford sue,” Opt. Express 13(6), 985–997 (2005). and New York (1999). 4. B. C. Wilson and G. Adam, “A Monte Carlo model for the absorption 30. T. J. Pfefer et al., “Light propagation in tissue during fluorescence spec- and flux distributions of light in tissue,” Med. Phys. 10(6), 824–830 troscopy with single-fiber probes,” IEEE J. Sel. Top. Quant. Electron. (1983). 7(6), 1004–1012 (2001). 5. S. T. Flock et al., “Monte-Carlo modeling of light-propagation in highly 31. C. G. Zhu and Q. Liu, “Validity of the semi-infinite tumor model in scattering tissues. I. Model predictions and comparison with diffusion- diffuse reflectance spectroscopy for epithelial cancer diagnosis: a Monte theory,” IEEE Trans. Biomed. Eng. 36(12), 1162–1168 (1989). Carlo study,” Opt. Express 19(18), 17799–17812 (2011). 6. S. E. Skipetrov and S. S. Chesnokov, “Analysis, by the Monte Carlo 32. D. J. Smithies and P. H. Butler, “Modeling the distribution of laser-light method, of the validity of the diffusion approximation in a study of in port-wine stains with the Monte-Carlo method,” Phys. Med. Biol. dynamic multiple scattering of light in randomly inhomogeneous 40(5), 701–731 (1995). media,” Quantum Electron. 28(8), 733–737 (1998). 33. G. W. Lucassen et al., “Light distributions in a port wine stain model 7. T. J. Farrell et al., “A diffusion theory model of spatially resolved, containing multiple cylindrical and curved blood vessels,” Lasers Surg. steady-state diffuse reflectance for the noninvasive determination of tis- Med. 18(4), 345–357 (1996). sue optical properties in vivo,” Med. Phys. 19(4), 879–888 (1992). 34. L. H. V. Wang et al., “Optimal beam size for light delivery to absorption- 8. L. H. Wang et al., “MCML-Monte-Carlo modeling of light transport in enhanced tumors buried in biological tissues and effect of multiple-beam multilayered tissues,” Comput. Meth. Prog. Bio. 47(2), 131–146 (1995). delivery: a Monte Carlo study,” Appl. Opt. 36(31), 8286–8291 (1997). 9. S. T. Flock et al., “Monte Carlo modeling of light propagation in highly 35. C. Zhu and Q. Liu, “Hybrid method for fast Monte Carlo simulation of scattering tissues. II. Comparison with measurements in phantoms,” diffuse reflectance from a multilayered tissue model with tumor-like IEEE Trans. Biomed. Eng. 36(12), 1169–1173 (1989). heterogeneities,” J. Biomed. Opt. 17(1), 010501 (2012). Journal of Biomedical Optics 050902-9 May 2013 • Vol. 18(5) Downloaded From: https://www.spiedigitallibrary.org/journals/Journal-of-Biomedical-Optics on 24 May 2021 Terms of Use: https://www.spiedigitallibrary.org/terms-of-use
You can also read